EEAT Mechanics

RankBrain twiddler: Google's first deep learning system, applied after scoring

Three schema objects suggest that part of RankBrain's output reaches the results as a twiddler, on a list that Ascorer, the main ranker, has already scored. RankBrain judges how well a page fits a query, not its quality, so this twiddler has no E-E-A-T bearing.

By Thomas Wawra· Published · Version 1.0· Systems referenced: RankBrain / DeepRank / FastSearch, Mustang / SuperRoot etc.

01What the schema shows

RESONEO lists RankBrain in its family of AI twiddlers, with three schema objects in the quality_twiddler namespace. RankBrainQueryFeatures holds features of the query. RankBrainTwiddlerResultSignals holds signals for each result. RankBrainTwiddlerResultAnnotation is an annotation, information the twiddler leaves on a result for later steps.

From these names RESONEO concludes that some of what RankBrain computes is not folded into the main score but applied afterwards in Superroot, the layer that fans a query out to Google's indexes and merges what comes back, on a list that has already been ranked. Like every twiddler, the RankBrain twiddler only recommends a move; the framework decides. How large the moves are and for which queries it acts is not known.

Source per claim (2)
C
RESONEO lists a RankBrain twiddler with three schema objects: RankBrainQueryFeatures (per query), RankBrainTwiddlerResultSignals (per result) and RankBrainTwiddlerResultAnnotation (information left for later steps).[1]
C
According to RESONEO, part of RankBrain's output is applied as a twiddler in Superroot, on a result list that Ascorer, the main ranker, has already scored.[2]

02What Google says about RankBrain

Google describes RankBrain as the start of deep learning in Search. It is a system that helps match queries to pages by their meaning rather than by their exact words. Our deep dive on RankBrain and DeepRank (/deep-dive/rankbrain-deeprank) covers what is known about it; this article does not repeat that.

What the schema adds is a place. RankBrain is usually discussed as part of core ranking. If RESONEO's reading is right, it has a second point of action after the score, where it can still adjust the order. The schema names do not show whether that second application uses the same model or a smaller one built for the twiddler.

Source per claim (2)
O
"When we launched RankBrain in 2015, it was the first deep learning system deployed in Search."[3]
C
If the schema names mean what they say, RankBrain acts in two places: in core ranking and again as a twiddler after scoring. Whether both use the same model is unknown.[4]

03E-E-A-T and the system map

On our map the RankBrain twiddler belongs to system 23, RankBrain and DeepRank. Those systems measure fit between query and page, not quality. A twiddler that carries RankBrain's output therefore has no direct E-E-A-T bearing: it can move a well-matched page up, but it does not judge who wrote it or how trustworthy it is.

For site owners there is no separate lever. What helps RankBrain in core ranking, a page that answers the need behind a query in plain language, is also the only thing that could help here.

Source per claim (1)
C
The RankBrain twiddler measures fit, not quality, and has no direct E-E-A-T bearing; there is no lever for site owners beyond what already helps RankBrain in core ranking.[4]

04References

This Deep Dive is Layer 2 content. It interprets the material and always points back to Layer 1, the reference layer. Every claim is mapped to a source, labeled by type: [A] Sworn testimony · [B] Verbatim from the leak · [O] Documented by Google · [W] Peer-reviewed study · [S] Checked on the page · [C] Our assessment · [P] Patent.

© Thomas Wawra · Senior SEO Manager · wetter.com — a Funke Digital company